In mid-May 2026, the Wikimedia Foundation fired Brooke Vibber, its first full-time employee and a former chief technology officer. One week later, it dissolved its six-person Community Technical Team. Vibber and many of the team’s engineers were union organizers. Days after the cuts, Jake Orlowitz, a former Foundation staffer who founded the Wikipedia Library, published a Medium essay titled “Big Tech’s Anti-Labor Playbook Has Come for Wikipedia” arguing that this is what it looks like: union busting. The Foundation denies the connection. The dispute has since moved from essay and petition to labor board process: the Foundation declined voluntary recognition in July 2026, and staff voted in September to form a union with the CWA.
The firings, one week apart
Vibber was the Foundation’s first full-time hire, MediaWiki’s former lead developer, and its CTO from 2005 to 2009. The Foundation had previously described Vibber as “one of the very few people in the world who have a deep understanding of the system’s technical foundations,” language that survives in Gigazine’s account of the dismissal.
Seven days later, the Foundation announced the dissolution of its Community Technical Team: five engineers and one manager. The team’s job was to translate requests submitted by editors through the Community Wishlist into shipped changes, which Gigazine’s account calls “effectively the only project where the volunteer community managed the product.” Dissolving it removes the primary channel through which editors could direct Foundation engineering resources toward their own priorities.
Vibber’s departure message was blunt. “Every worker has the right to have their voice heard in how the workplace is run, and every workplace needs a union,” the message read, calling on all Foundation employees to participate in the unionization effort. Vibber also stated an intention to keep contributing as a volunteer and to continue developing the open-source portion of MediaWiki, which shifts some of the deepest knowledge of the platform off payroll.
The Foundation’s account
The Foundation rejects the union-busting reading. Executives said the Community Wishlist would be handled by other teams, that the organization’s actions were unrelated to union activity, and that the affected employees were being evaluated for other positions within the organization, according to Wikipedia’s summary of the episode. The editing community noticed a detail that strains the placement promise: the Foundation listed fewer developer job openings than the number of staff laid off.
Two facts complicate both sides. The cut landed under a new chief executive: Bernadette Meehan took office on January 20, 2026, succeeding Maryana Iskander, roughly four months before the dismissals. And the denial is narrow. It addresses motive, not effect. Even granting that the disbanding was unrelated to union activity, the result is the one Orlowitz describes: the staff most active in the union and the team that gave editors a say over product direction are both gone, and the burden of explaining the timing now sits with management.
The critic
Orlowitz is not a random agitator. He joined the Wikimedia Foundation in 2014 as a senior program manager and left in 2019. He founded the Wikipedia Library, which partnered with JSTOR, Oxford University Press, Elsevier and over a dozen other publishers or platforms to give top contributors free research access; by his own record, that meant free access to 100,000 academic journals for 25,000 editors, and 70 long-term relationships with scholarly publishers. He also built The Wikipedia Adventure, an onboarding game for new editors. His political-theory background (Wesleyan) shows in how he frames the dispute: not as a personnel grievance but as a structural failure of institutional governance.
His argument, as Gigazine recounts it, is that the Foundation has lost community trust by making top-down decisions “like a Silicon Valley IT company for nearly the past decade,” and that the firings follow a recognizable industry script: “They fire the engineers who understand how the system works, they fire the people who organize unions, and they pray that a catastrophic problem doesn’t occur before they ship out a fancy product. Twitter, Meta, Salesforce, Google, they’ve all done this.”
The comparison is pointed because it names real companies. Whether that script describes a documented industry-wide pattern is Orlowitz’s assertion; the sources here record his argument, not an independent history of platform restructuring. What they do establish is the difference in business model: those companies sell ads and cloud contracts. The Wikimedia Foundation runs on small-donor fundraising banners and, increasingly, AI licensing revenue.
The AI licensing backdrop
In January 2026, the Wikimedia Foundation signed AI training deals with Amazon, Microsoft, Meta, Perplexity, and Mistral AI through its Enterprise program, which provides high-volume, high-speed access to over 65 million articles. Google had signed a similar deal in 2022, and smaller firms like Ecosia followed. Financial terms were not disclosed.
Jimmy Wales acknowledged the tension directly: “They’re not donating in order to subsidize these huge AI companies.” He supported the deals on quality grounds, telling the AP he was “very happy personally that AI models are training on Wikipedia data because it’s human curated.”
The undisclosed terms invite assumptions of windfall money, so it is worth grounding the figure. The Foundation’s FY2024-2025 Enterprise report put Enterprise revenue at about $8.3 million, up 148 percent year over year, roughly 4 percent of total revenue, and the program’s first full year of profitability. Against the Foundation’s total FY2024-2025 revenue of $208.6 million, licensing is a fast-growing but minor contributor, not the budget. That cuts against the crude story that licensing cash bought the layoffs, and it sharpens the structural one: the dispute is about institutional posture toward contributors and AI buyers, not a single revenue line large enough to drive a restructuring on its own.
The deals are not the cause of the firings. But the timing is hard to ignore: the Foundation is collecting undisclosed sums from five of the largest AI companies on earth, while eliminating the team that gave volunteer editors their main lever over product direction, and firing the CTO-era engineer who was organizing a union. Whether or not there is a causal line between the revenue and the layoffs, the optics match the structural pattern Orlowitz describes: institutional monetization compresses contributor agency.
Why the AI deals exist at all: scrapers were already running up the bill
The licensing program did not appear in a vacuum. AI companies were taking the data whether or not anyone paid, and the cost showed up in Wikimedia’s own infrastructure reporting. In an April 2025 engineering post, the Foundation reported that bandwidth for multimedia downloads had grown 50 percent since January 2024, driven largely by automated scrapers harvesting the Wikimedia Commons image catalog to feed AI models rather than by human readers. The skew was sharp: at least 65 percent of the most expensive core-datacenter traffic came from bots, while bots generated only about 35 percent of pageviews, because crawlers bulk-read the unpopular pages that human-optimized caching never warms. The Foundation’s response was a “responsible use of infrastructure” workstream for the following fiscal year, and in April 2026 it updated the post to point to a one-year follow-up, “Quo Vadis, Crawlers? Progress and what’s next on safeguarding our infrastructure.”
That is the backdrop the Enterprise deals answer. A paid, rate-limited API that serves machine clients a clean dump is cheaper to run than absorbing an open-ended crawl, and it converts a cost center into a revenue line. The same dynamic is playing out across the web, where infrastructure providers now meter and price crawler access rather than eat it, and even the size of the surge is contested; whether Cloudflare’s bot-traffic numbers are real is its own measurement dispute. The load has also shown it can break things. A major Wikidata outage in May 2026 was attributed by the Foundation on October 6 to rogue OpenAI agents that, in its account, were sending thousands of queries, altering Wikimedia’s citation tool and attempting to compromise Etherpad, with OpenAI saying it would work with the Foundation to investigate, according to Wikipedia’s summary. For Wikimedia the deals are defensible engineering. The problem Orlowitz raises is what happens to the humans inside the institution while it optimizes its relationship with the machines outside it.
The structural problem: the Wishlist was the last community-managed product
The Community Technical Team was not a support function. It was the only Foundation engineering group where the volunteer community set priorities through a structured process: editors submitted requests to the Community Wishlist, and the team evaluated, prioritized, and shipped them. Gigazine’s description of it as “effectively the only project where the volunteer community managed the product” is the kind of claim that is either true or trivially falsifiable, and nothing in the reporting here contradicts it.
Dissolving the team does not just eliminate six jobs. It eliminates the mechanism by which the people who produce Wikipedia’s content, without pay, could direct the Foundation’s technical resources. The editors write the articles. The Foundation collects the donations and, now, the licensing fees. The Wishlist was the channel where editorial labor translated into institutional response. That channel is gone.
This is the same dynamic that Ford et al. (2024) identified in their research manifesto on Wikimedia as public knowledge infrastructure: AI tools have made Wikimedia content foundational to circulating knowledge, raising questions about “the agency of Wikimedia contributors, and the outcomes of their labour.” Their first stated principle for researchers is to “map the dispossession of the commons.” The manifesto framed this as a research agenda. The firings are an empirical data point.
The nonprofit-platform convergence
Orlowitz’s core claim is that the Foundation’s 20-year message, that “we are different from other technology companies,” will be judged by how it handles the union drive. The fact that baseline workplace governance requires a union campaign to advance inside a donate-funded nonprofit is itself the signal he is pointing at.
The convergence shows up most clearly in how institutions route contributor output to AI buyers. The pattern is rarely a lawsuit or a loud announcement: it is a default setting, AI training on unless the people who made the content go looking for the opt-out (see our related coverage of AI-training defaults at Atlassian). Wikimedia’s posture is more defensible, since its content is already openly licensed and the Enterprise deals are negotiated rather than silent, but the structural move is similar: the entity that hosts the contributors’ work decides how that work is monetized for AI, and the contributors learn the terms after the fact. The difference Orlowitz presses is governance. A for-profit can set defaults unilaterally and call it a product decision. A nonprofit that exists to steward a commons is supposed to answer to the people who build it.
The second-order risk is specific to Wikipedia’s production model. For-profit platforms that degrade worker conditions face churn, quality decline, and competitive pressure. The workers leave. For Wikipedia, the workers are unpaid volunteers who contribute because they believe in the project’s mission. If the institution that administers that mission loses their trust, the contributors do not leave for a competitor. They stop contributing. The commons degrades, and the AI companies that paid for access to a curated knowledge base find the curation deteriorating.
What happens now
Editors moved first. A solidarity page on English Wikipedia affirms a willingness to engage in collective action “up to and including staging an editorial strike,” and the associated petition had drawn 1,190 signatures by July 24, 2026, according to Wikipedia’s account. The strike under discussion was specific: stopping voluntary editing, the one labor input the Foundation cannot replace with a hire.
The union then forced the legal question. On June 24, UK-based Foundation employees formally requested recognition through the United Tech and Allied Workers section of the Communication Workers Union, a move CWU spokesperson John Chadfield described as “the first attempt of Wikimedia Foundation workers to seek voluntary recognition anywhere.” On July 7, the Foundation and its UK employer of record, Safeguard Global, refused voluntary recognition but agreed to negotiate within a 20-day window, according to the community-maintained timeline. On July 20, US-based employees requested recognition of Wiki Workers United’s US branch, backed by the Communications Workers of America, with a July 24 deadline for a response. On July 27, the Foundation declined voluntary recognition, saying it would respect the results of a secret-ballot vote organized by the National Labor Relations Board, which it viewed as “the appropriate path forward.” 404 Media, as quoted in Wikipedia’s coverage, described that statement as “very carefully-worded language common among companies and organizations that have fought against unionization.” The NLRB process was registered on July 28, listing a Littler Mendelson representative for the Foundation and Weinberg, Roger & Rosenfeld for the CWA.
On September 4, 2026, Wiki Workers United announced that Foundation workers had “overwhelmingly voted to form a union with the CWA.” Recognition and a first contract are separate fights, and neither outcome is settled. What has changed is that the dispute now runs through an NLRB case and a documented vote rather than an essay and a petition.
The structural question is not whether the Foundation had the legal right to restructure. Organizations restructure. The question is whether a nonprofit whose sole asset is the unpaid labor of roughly 277,000 volunteers can afford to treat its staff the way for-profit platforms treat theirs, and whether the Enterprise revenue stream creates the same institutional incentives that degrade contributor trust at every other platform company that has walked this path. Orlowitz’s argument is that the answer is already visible in the pattern. The burden of proof has shifted.
Join the discussion
Share a useful perspective or ask a question about this article.